Introduction
# A tibble: 823 × 6
track.name tempo loudn…¹ energy tempo2 tempo3
<chr> <dbl> <dbl> <dbl> <dbl> <dbl>
1 Colors - Radio Edit 150. -3.40 0.937 150. 150.
2 Lessons In Love - Headhunterz Remix Radio… 150. -3.86 0.893 150. 150.
3 Never Say Goodbye - Wildstylez Radio Edit 148. -3.41 0.834 148. 148.
4 Year Of Summer - Radio Edit 150. -5.57 0.888 150. 150.
5 Catch Me (feat. Naaz) 145. -3.33 0.916 145. 145.
6 Rockstar (feat. DV8) 128. -5.75 0.973 128. 166.
7 Our Church 150. -3.47 0.957 150. 150.
8 Destiny - Edit 150. -4.63 0.791 150. 150.
9 Home 150. -3.81 0.896 150. 150.
10 Waiting For You - Rebourne Remix 75.0 -3.34 0.974 150. 150.
# … with 813 more rows, and abbreviated variable name ¹loudnessIn this storyboard I will attempt to explain the differences between Hardstyle and Hardcore. The data I will use for this project is my own playlist called “Kasper Hardstyle/core”. In this playlist you can find a combination of the two mentioned genres. There are also some songs which could be categorized in the genre Frenchcore but the percentage of songs that could be identified with that genre is very small in my playlist. The reason I chose this subject is because I am very passionate about these genres. Especially since in my spare time I produce songs in these genres. I will try to create a clear view of the differences of these genres and the first difference I’d like to highlight is the difference in tempo. Hardstyle usually has a bpm ranging from 140 to 165, while Hardcore is a bit quicker, most songs are between the 180 and 220 songs. The differences in bpm in both genres can be explained because of the many subgenres that both genres have. The more melodic or “euphoric” subgenres tend to be slower and less melodic subgenres have faster tempo’s.
In the plot you can see tempo on the x-axis with loudness on the y-axis the tempo, the colour of the data points are determined by energy of the tracks. With this plot I’m trying to see if you can clearly see a difference in the genres when it comes to loudness and if you can really separate the two by just tempo. Both these things are the case. The line in the plot is slowly rising as the bpm increases which means that the tracks are gaining loudness on average. The reason this is interesting is because loudness in the case of these songs can also be interpreted as harshness. The songs that are all the way on the right are normally regarded as very rough and harsh songs. The other fun thing is that around the 160 bpm you can very clearly see that the songs switch from genre.
Unexpected chromagram result
The song I selected for analyzing the chroma is “Warriors 2022 edit” by DRS. The reason I selected this song is because I expected it to have very interesting results since it has sections with only one sound playing. These sounds are in the case of Hardcore and Hardstyle nearly always the kicks. In this song you can see some very clear brightness between 5 seconds and 60 seconds. These bright spots are the vocals of the song. That is the way to recognize the introduction. I expected to see more clear sections in the chromagram, but after listening to the song again while analyzing the chromagram I realized why the sections flow more nicely into eachother than I expected. The reason is that the genres I’m covering in this storyboard tend to use multiple choruses that flow into each other. But if you look closely you can see where the choruses are. An example would be to look at the timeframe 100 seconds to roughly 110 second. In that time you see a brighter bar in the E and F notes. This is because those are the two notes that the kick uses for that duration. So that is one of the sections of the song. By doing the same and looking for bars of lighter colours you can spot the sections of the song. This however is far from what I expected since I expected the chromagram to be emptier with more bright bars at the choruses.
Selfsimilarity matrices part 1
In this part of the storyboard I decided to compare the songs “AEON” By DEEZL and “Ignition” by Spitnoise and Deadly Guns. The type of comparison I’ll be doing is a comparison of a selfsimilarity matrix of both songs. I selected these songs because they fill a similar spot in the genres. Both songs have a very rhythmic first chorus and a more melodic last chorus with returning elements and melodies from the verses. I decided to use timbre for the selfsimilarity matrices because I expected that to return the most interesting results and nice contrast between the genres. Furthermore, the matrices are displayed in beats since that yielded the best results. So let look at the results of both matrices.
The first thing to notice when looking at the matrix of “AEON” is that there is a lot of repetition in the song. There is most noticeable a checkerboard pattern around the 20 seconds, 60 seconds, 125 seconds and in the outro of the song. These checkerboards is the arpegiating melody that plays throughout the entire song. The second very noticeable thing is the bright cross at 40 seconds. This bright cross is the first chorus. As you can see, it is hard to find the structure of the song in this matrix. This is especially the case since the same melody is comprised of the same sounds which played in the verse and the chorus. Now let’s look if these findings are also noticeable in the Hardcore song.
Selfsimilarity matrices part 2
The first thing that struck me when I looked at this matrix is that it seems a lot more structured. So let’s see if that is really the case. This matrix has a lot more apparent boxes and squares in it. In these squares you can see more squares but all the bigger squares seem different from the others. This makes it seem as if in this case you can clearly see the different parts of the song. After listening to the song again it does really seem like the matrix picks up the different sections of the song. The first square which lasts until roughly 70 seconds is the introduction and the first chorus. These sections use a lot of similar sounds and are quite empty. Then after that until 150 seconds, the first verse can be found which transitions into a second chorus with similar sounds again. At last there is a verse and a chorus, however, these are quite different which can be seen by looking at the pattern transition at roughly 170 seconds. Although the verses and choruses are very hard to separate in this matrix, it does represent the changing sounds and timbres of the song nicely.
Euphoric keygram
The song I used for this keygram analysis is “Together We Grow” by “Vertile”. I wanted to highlight a different side of the genres I selected by looking at a very melodic track which is coincidentally one of my favourite songs. What I expected to see was a very constant pattern since the melody of the song keeps being repeated. The result however surprised me a bit. Even though the melody does not change in the song and the instruments that play the melody also don’t (drastically) change, there are very clearly highlighted areas. Explaining why the graph looks like this is relatively tough. Even though it gets the key right because the darkest line going through the keygram is f sharp, it’s still just marginally darker than the other visible lines in the graph. Furthermore, the pattern in the graph is hard to explain. For example, the dark streak in the beginning of the graph is a more melodious bit with nearly just the melody playing. But the very dark patch at the end of the graph is where the melody mostly stops playing, so how distinct the melody is doesn’t factor in. My final theory for why the algorithm struggles so much to figure out which key it is, is because the melody has been very heavily processed with things such as delay, reverb and the sound itself is not a very conventional synthesizer sound in most genres. The type of synthesizer that is used is called a super saw since it is a lot of saw waves layered on top of each other. So in conclusion, I tried to show a more emotional and melodic side of Hardstyle which isn’t often found in Hardcore, but the graph ended up showing a very interesting result which doesn’t help with highlighting the differences between the genres. But it does create opportunities for experimental thinking why the graph looks the way it does and thinking about how to improve the current systems.
Tempo
The track that is displayed in the tempogram on this page is Homicide by Rejecta and Act of Rage. The reason I selected this track is because I wanted to see if the algorithms behind the tempogram could accurately predict the tempo when a track contains “kickrolls”. A kickroll is a pattern made with the kick of the track in the verse which deviates from the standard 4/4 pattern in the measure. I knew that the system could probably accurately predict the tempo of a hardcore track since it has a more steady beat and doesn’t contain as many kickrolls. Therefore I selected this track ,because this is a hardstyle track. Since kickrolls usually don’t fall on the 4 beats in a measure I expected this to be interesting. If you look at the tempogram you can see that it predicts that the song is mainly around the 480 bpm. This isn’t strange since that bpm is two tempo octaves above the actual bpm. You can also see a lot of bright points deviating from the straight line in the choruses and verses. These are mainly the kickrolls and this was what I was hoping to see. Not all kickrolls are visible but one of the most visible ones is around the 60 seconds. Furthermore it is interesting to see that the algorithm recorded the tempo going downward around the 45 seconds where that is actually a synth riff. The last thing that is noteworthy is that turning on the cyclic mode of the tempogram actually made the graph worse, usually it would make the graph more readable.
Using AI
In this bit of the storyboard we will be taking a step back and look at the larger picture. The purpose of this storyboard is to find differences and similarities between the two genres I picked. So I decided to use the random forests algorithm to see if AI is able to spot the differences between the genres. The parameters I trained the algorithm on are the following: danceability, energy, loudness, valence and tempo. The reason I picked those are because I felt like the other parameters didn’t matter much for my selection of genres. I had to deal with an issue, that being that my normal corpus is too big and doesn’t have labels for which genre each song is. So I used randomization to pick songs for me and I labeled those by adding them to a hardstyle and hardcore playlist. In the end the algorithm actually does this very well. It has a precision of 0.88… for hardcore and 1.00 for hardstyle. The recall for hardcore is 1.00 and for hardstyle it is 0.95. These are Obviously very good scores. In the plot I added you can see that tempo is a very big difference for the two genres, I therefore think that the algorithm mostly classifies by tempo.
Conclusion and Discussion
In this last section I will write of the findings I made while analyzing my corpus. The most notable finding is that a very good way to categorize Hardstyle and Hardcore is their respective BPM’s. I always knew that Hardcore was faster but I really thought that there was more nuance to it than just saying that Hardcore is always 180 BPM or higher and Hardstyle stays at 160 BPM or bellow. Of course there are also theme and sound differences like different synthesizer selection, but that isn’t as defining. Especially since in that regard the two genres do have overlap.
Furthermore, it was very interesting to see current analyzing techniques struggle with these genres. A good example would be the chromagram results I found. I do now have a deeper appreciation for how hard it is to analyze music, especially different types of music.
The last things that I want to mention are a couple of things that could have been very interesting to do with more time if I wanted to go even deeper into the differences of these genres. The first things would be to go very deep into the sounds used and the way they are characteristic of their respective genres. Something that I would improve upon if I continued working on this portfolio is that I would put more time into a more thorough classification test using multiple different AI algorithms.
I hope that you enjoyed going through my portfolio, my goal has been to highlight music styles I love and to share it with people. The best result would be to create a bit of interest in the reader because in my opinion these genres are very misunderstood.